1. Field of the Invention
The present invention relates to a load dispersion system, a control method for a load dispersion system, and a storage medium.
2. Description of the Related Art
As a mode for conducting various types of processing on a computer, the technology of a cloud computing system or SaaS (Software as a Service) has been proposed. With cloud computing, it is possible to simultaneously process requests from many clients by utilizing numerous computing resources, and by conducting dispersed execution of data conversion and data processing. In order to make good use of the features of cloud computing, systems have been proposed in which numerous jobs are scalably processed by having a server carry out a series of processing tasks on request, and process these simultaneously and in parallel. Such a system is hereinafter referred to as a cloud system.
With respect to construction of a cloud system, it is necessary to have a configuration in which multiple servers conduct load dispersion of requests. For example, the information processing system disclosed in Japanese Patent Laid-Open No. 2007-164527 is a system provided with a load dispersion apparatus which receives a request from a user utilizing a web service, and which decides upon one of multiple web servers, thereby dispersing the load imposed on the servers. A load dispersion apparatus is also referred to as an SLB (software load balancer). An SLB is set up to mediate communication between communication-source and communication-target server groups, and disperses communications from the communication sources among multiple servers based on the (TCP/UDP) port numbers of the communication application.
With the system disclosed in Japanese Patent Laid-Open No. 2007-164527, user requests are transmitted to servers via a load dispersion apparatus, but when the SLB is transited, network bandwidth is restricted by the upper limit of SLB processing performance. In some cases, when the SLB is transited, processing speed is half or less of network bandwidth compared to the case where the SLB is not transited. Therefore, with systems which conduct exchanges of relatively large volumes of binary data such as documents and images, the SLB constitutes a performance bottleneck. As a result, extra time is required in order to process user requests.